In voice recognition for the conventional car navigation system, for example, a speaker presses down a talking switch before starting a speech, which silences the car navigation system, preventing the sound output by a loudspeaker of the car navigation system from being transferred to a microphone as echo, which may be a cause of false voice recognition. Users, however, desire to be able to operate by voice recognition without silencing of the output sound. To meet such a request, an echo suppressing method for suppressing the sound transferred from a loudspeaker to a microphone is required. Moreover, since a plurality of speakers are used for audio in a car, i.e. a car audio system, to respond to a plurality of channels, an echo suppressing method which addresses echo transferred from a multichannel car audio system to a microphone used for voice recognition or the like is desired.
According to the Japanese Laid-open Patent Publication No. 2002-237769, an echo suppressor has been proposed in which an echo suppressing method that addresses a monaural channel audio system is applied to a multichannel audio system.
However, the echo suppressing method requires a suppression mechanism corresponding to each channel and increases in cost and size of a device. The problem of size is serious especially when the device is applied to a car navigation system for which the installation space is largely limited.
Moreover, when reference sound signals are added up to obtain a summed reference sound signal of a monaural channel, a problem occurs such that unsuppressed residual echo becomes larger. This is because the reproduced sound and intensity of each loudspeaker varies independently in the output of sound such as music by the multichannel audio apparatus, which degrades presumption accuracy if echo from a plurality paths are learned by one adaptive processing.